Last updated October 3, 2026
The ISTQB Certified Tester Testing with Generative AI exam (CT-GenAI v1.1) has 40 questions worth 46 points, runs for 60 minutes, and needs 30 points to pass — 65%. CTFL is a prerequisite, the certificate never expires, and the questions are drawn from five syllabus chapters in a published split of 7, 11, 10, 5 and 7.
Scoring is weighted. A K1 or K2 question is worth one point and a K3 question is worth two, which is how 40 questions add up to 46 points. That weighting is the single most misread fact about this exam: 65% means 30 of 46 points, not 26 of 40 questions. Six correct K3 answers are worth as much as twelve correct K1 answers, so a candidate who is strong on definitions and weak on application can clear 26 questions and still miss the mark.
Questions: 40
Points: 46 (K1 and K2 = 1 point, K3 = 2 points)
Pass mark: 30 points, which is 65%
Time: 60 minutes, or 75 minutes with the standard 25% allowance for sitting in a non-native language
Prerequisite: a valid CTFL certificate
Syllabus version: v1.1, in force since 27 April 2026 (v1.0 was released 25 July 2025)
Validity: the certificate is issued for life and needs no renewal
Sixty minutes for 40 questions is 90 seconds each on average, and the exam is booked through an ISTQB member board or an accredited exam provider, which is also who decides whether you sit it remotely or in a test centre.
This is the question most people arrive with, and the answer is clean. CT-AI is about testing AI-based systems: training data, model quality characteristics, functional performance metrics, bias, explainability. CT-GenAI is about using generative AI in your own testing work: writing prompts that produce usable test conditions, judging what the model gives back, and managing the risk that comes with it.
Neither exam is a prerequisite for the other and both sit directly on CTFL. Sit CT-GenAI if an assistant has already become part of how you do test analysis, test data or review work and you want the practice to be disciplined. Sit CT-AI if the product under test is itself built on machine learning. Plenty of testers end up with both, and the longer comparison lives in our article on CT-GenAI versus CT-AI.
ISTQB labels every learning objective with a cognitive level. K1 is recall — name it. K2 is understanding — explain it, compare it, say why. K3 is application — do it on a given situation. CT-GenAI has no K4 questions at all, so nothing on this exam asks you to analyse a scenario and decompose it the way an Advanced Level paper would.
The official split is 8 K1, 26 K2 and 6 K3. Those six K3 questions carry 12 of the 46 points: 15% of the questions, 26% of the marks. If a practice result shows solid K1 and K2 with the K3 questions missed, the diagnosis is specific — you know the vocabulary and you cannot yet apply it to a prompt or a risk decision. That gap is closed by writing prompts, not by rereading the chapter. Our explainer on ISTQB K-levels covers how the weighting works across the other exams.
The per-chapter distribution is published, not inferred, and every mock exam on this site is bound to it exactly. Read the running total as a budget: it shows where points are still available and where they have already run out.
Introduction to GenAI for Software Testing — 7 questions, 17.5% of the exam, 7 points. Running total: 7 of 46.
Prompt Engineering for Effective Software Testing — 11 questions, 27.5% of the exam, 16 points, because five of the six K3 questions live here. Running total: 23 of 46.
Managing Risks of GenAI in Software Testing — 10 questions, 25% of the exam, 11 points (one K3 sits here). Running total: 34 of 46.
LLM-Powered Solutions for Software Testing — 5 questions, 12.5% of the exam, 5 points, all K2. Running total: 39 of 46.
Deploying and Integrating GenAI in Test Organizations — 7 questions, 17.5% of the exam, 7 points. Running total: 46 of 46.
Two consequences are worth sitting with. The first three chapters carry 34 points between them, and the pass mark is 30 — so a candidate who is genuinely strong on generative AI fundamentals, prompting and risk has the exam in hand before the last two chapters are even reached. The mirror image is harsher: lose all of Chapter 2 and the remaining ceiling is exactly 30 points, the pass mark itself, with no margin for a single mistake anywhere else. The syllabus agrees with the exam here — Chapter 2 is also the largest training block, 365 of the 815 minutes of accredited training the syllabus requires.
Read the syllabus end to end once, at speed, to see the shape of it. Then stop treating the chapters as equal. Work Chapter 2 until a structured prompt is something you write rather than recognise: role, context, task, constraints, output format, and an explicit statement of what the model should do when it does not know. The K3 objectives are about applying prompting to real test activities — deriving test conditions, generating test data, drafting a review checklist — so practise on your own backlog, not on toy examples.
A workable order after that is Chapter 3, then 1, then 5, then 4: risk is the second-largest block and the one where careless reading costs most, while Chapter 4 is five K2 questions and does not reward deep study. Use random practice from the question pool to drill one chapter at a time, then switch to full mocks: Mock 1, Mock 2 and Mock 3, each one holding the official 7/11/10/5/7 distribution so a practice score means the same thing a real score would.
Book the exam when three full mocks land at or above 30 points and the per-chapter breakdown shows no chapter sitting below its target twice in a row. The breakdown after each attempt splits your result by chapter and by K-level, which is what turns a score into a study decision, and every answer option carries a written rationale — including the wrong ones, which is where the misconception you actually hold is explained. Every mock is free and needs no sign-up; an account exists only to keep your attempt history, and results can be exported to PDF.
Reading the pass mark as 26 of 40 questions. It is 30 of 46 points, and the K3 weighting is what separates the two.
Revising CT-AI material by mistake. Training data quality, precision and recall, model drift — none of that is examined here.
Treating prompt engineering as vocabulary. Knowing what chain-of-thought means answers a K2 question; a K3 question asks you to choose and shape a prompt for a stated test activity.
Memorising vendor and product names. The syllabus is deliberately vendor-neutral, and an option that names a specific tool is usually there as a distractor.
Skipping Chapter 5 as management padding. It is 7 questions, the same weight as Chapter 1, and the easiest 7 points on the paper to collect.
Reading the risk chapter as law. It is 10 questions about the risks of using generative AI in testing — hallucination, leakage, over-reliance, provider due diligence — not a regulatory summary to memorise.
Losing the clock. At 90 seconds a question there is no room to re-read a long stem three times; flag it and come back.
40 questions · 46 points · 30 to pass (65%) · 60 minutes (75 non-native)
Question weights: K1 = 1, K2 = 1, K3 = 2, no K4
K-level split: 8 K1 · 26 K2 · 6 K3
Chapter split: 7 · 11 · 10 · 5 · 7 questions; 7 · 16 · 11 · 5 · 7 points
Heaviest chapter: Prompt Engineering, 16 of 46 points from 11 questions
Prerequisite: CTFL · Training: minimum 815 minutes accredited · Certificate: for life
Syllabus v1.1 in force since 27 April 2026
Shorter answers to the questions people ask before booking are on the CT-GenAI FAQ page, and the official exam details and syllabus download are on the ISTQB certification page. Exam rules and the K-level definitions themselves are published on istqb.org/help/exam.